The amount of resolution you need in an image is dependent on three factors. One is the amount of enlargement as described above. Second, we need to know the frequency of the halftone screen (even stochastic patterns have a screen frequency; the cells are still present-even if the dots aren't). Screen frequencies range from coarse for posters and large displays (usually about 35 lines per inch) up to high quality sheet-fed printing at 200 Ipi or more. Determining the correct frequency is a function of choosing a printing process and then determining the appropriate screen frequency for that process. Printers of all processes will be able to tell you the limitations of screen frequency on their presses. Newsprint, for example, is usually printed with screen frequencies of 85 or 100 lines per inch. Web-fed magazines are usually printed with 133 lines per inch, and commercial printing on sheet-fed presses is usually printed at 150 lines per inch. We choose the correct one for the calculation of resolution, and proceed with our formula.

Next, we need a multiplier. The multiplier is used to generate more than one pixel in the image per eventual halftone dot in the printed halftone. Conventional wisdom" has recently dictated that a multiplier of 2.0 is correct, and many people adhere to this value as if it were a religious truth. But, experimentation will show that most images don't need this much resolution. Practical values range from 1.25 to 2.0 for conventional dot halftones, and multipliers of less thus 1:1 suffice for the new stochastic pattern images. The multiplier is chosen in part according to the halftone frequency and content of the image Successful stochastic printing is now being done from images scanned (or, in the case of Photo CD scans, chosen) at much lower relationships between the pixels and 'dots" - as low as 0.75:1 and lower.

A common engineering cornerstone, Nyquist's theorem, dictates that to get a good sample you must scan 2:1 data for final analysis or any analysis of information. 4'Nyquist's theorem," in the words of one scientist I interviewed, "was an emancipation for engineers. It told us that we didn't need more than 2:1 data ratio on scans; it freed us to sample, and get on with the analysis without worrying about whether we were introducing error into the study from too little data." Nyquist's theorem, however, has more to do with signal-to-noise ratio in transmission and engineering analysis than it does in halftones because of the random nature of continuous tone images. Errors introduced by sampling at lower ratios result in nothing more than a slight loss of edge sharpness. Using unsharp masking, one can easily restore a sense of sharpness to an image undersampled for reproduction. Fine detail can be lost, however.

So, we can experiment with sampling ratios of 1.25 or 1.5 to complete our formula. You will see quickly that image usage is a flexible, virtually elastic, issue when it comes to reproduction, and that no absolute rules apply. What occurs in halftone-making is the passing of a digit from one process to another, with under- or oversamplings taken from point to point, for the ultimate of an averaging of data to create a tonal dot or pattern to impart the illusion of print tonality. The entire halftone process creates an illusion of tonality where none exists.

Next
Colorite Home Page
OPTIMIZING PHOTO CD -- MENU